WO2021182041A1 - Abnormality determination method for blast furnace, training method for stabilization period model, operation method for blast furnace, and abnormality determination device for blast furnace - Google Patents

Abnormality determination method for blast furnace, training method for stabilization period model, operation method for blast furnace, and abnormality determination device for blast furnace Download PDF

Info

Publication number
WO2021182041A1
WO2021182041A1 PCT/JP2021/005879 JP2021005879W WO2021182041A1 WO 2021182041 A1 WO2021182041 A1 WO 2021182041A1 JP 2021005879 W JP2021005879 W JP 2021005879W WO 2021182041 A1 WO2021182041 A1 WO 2021182041A1
Authority
WO
WIPO (PCT)
Prior art keywords
blast furnace
value
stable period
abnormality
abnormality determination
Prior art date
Application number
PCT/JP2021/005879
Other languages
French (fr)
Japanese (ja)
Inventor
島本　拓幸
啓史 小橋
Original Assignee
Ｊｆｅスチール株式会社
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ｊｆｅスチール株式会社 filed Critical Ｊｆｅスチール株式会社
Priority to JP2021533830A priority Critical patent/JP7192992B2/en
Publication of WO2021182041A1 publication Critical patent/WO2021182041A1/en

Links

Images

Classifications

    • C—CHEMISTRY; METALLURGY
    • C21—METALLURGY OF IRON
    • C21B—MANUFACTURE OF IRON OR STEEL
    • C21B7/00—Blast furnaces
    • C21B7/24—Test rods or other checking devices

Definitions

  • the present invention relates to a blast furnace abnormality determination method, a stable period model learning method, a blast furnace operation method, and a blast furnace abnormality determination device.
  • Patent Document 1 proposes a technique for grasping a sign of an operation abnormality by using a Q statistic.
  • the present invention has been made in view of the above, and learning of a blast furnace abnormality determination method and a stable period model capable of detecting an operation abnormality of a blast furnace at an early stage by simultaneously using various data having different characteristics. It is an object of the present invention to provide a method, a method of operating a blast furnace, and an abnormality determination device for a blast furnace.
  • the blast furnace abnormality determination method uses a plurality of operation data in the stable period of the blast furnace so that the input value and the output value are the same.
  • a plurality of operation data to be determined are input to the learned stable period model, and an abnormality determination step of determining an operation abnormality of the blast furnace based on the difference between the input value and the output value at that time is included. ..
  • the abnormality determination step is each input value and each output when a plurality of operation data to be determined are input to the stable period model.
  • the integrated value of the difference from the value is calculated, and when the integrated value of the difference exceeds a preset threshold value, it is determined that there is an operation abnormality.
  • each input value and each output when the abnormality determination step inputs a plurality of operation data to be determined for the stable period model is calculated for each positive and negative, and when the integrated value of the positive side difference exceeds the preset positive side threshold value, or the integrated value of the negative side difference is preset. When the threshold on the negative side is exceeded, it is determined that there is an operation abnormality.
  • each input value and each input value when a plurality of operation data to be determined are input to the stable period model after the abnormality determination step. It further includes a display step of displaying the integrated value of the difference from the output value for each positive or negative in a stacked graph.
  • the learning method of the stable period model according to the present invention includes input values and output values by inputting a plurality of operation data in the stable period of the blast furnace to the autoencoder. Includes learning steps to build a stable model trained to be the same.
  • the learning step selects a plurality of operation data in the stable period of the blast furnace based on the air flow rate, and constructs the stable period model. ..
  • the learning step is such that the blast flow rate is equal to or more than a preset threshold value among a plurality of operation data of the blast furnace, and the blast flow rate is the same.
  • the operation data excluding the predetermined time before and after the time when is equal to or more than the threshold value is selected, and the stable period model is constructed.
  • the blast furnace operation method changes the operation of the blast furnace based on the determination result of the above-mentioned blast furnace abnormality determination method.
  • the blast furnace abnormality determination device uses a plurality of operation data in the stable period of the blast furnace so that the input value and the output value are the same.
  • a plurality of operation data to be determined are input to the learned stable period model, and an abnormality determination means for determining an operation abnormality of the blast furnace is provided based on the difference between the input value and the output value at that time. ..
  • the present invention by performing anomaly determination using a stable period model trained so that the input value and the output value are the same, various data having different characteristics can be used simultaneously to determine the operation abnormality of the blast furnace. Detection can be performed early.
  • FIG. 1 is a diagram showing a schematic configuration of an abnormality determination device for a blast furnace according to an embodiment of the present invention.
  • FIG. 2 is a diagram showing an outline of an autoencoder used in the learning method of the stable period model according to the embodiment of the present invention.
  • FIG. 3 is a flowchart showing the flow of the learning method of the stable period model according to the embodiment of the present invention.
  • FIG. 4 is a flowchart showing the flow of the abnormality determination method according to the embodiment of the present invention.
  • FIG. 5 is an example of an abnormality determination method for a blast furnace according to the present invention, and is a diagram showing an example in which an abnormality determination is performed using verification data for a predetermined period.
  • the blast furnace abnormality determination method, the learning method of the stable period model, the blast furnace operation method, and the blast furnace abnormality determination device will be described with reference to the drawings.
  • the blast furnace abnormality determination device, the learning method of the stable period model, the blast furnace abnormality determination method, and the blast furnace operation method will be described in this order.
  • the present invention is not limited to the embodiments described below.
  • the abnormality determination device 1 is for determining an abnormality in a plant such as a blast furnace. As shown in FIG. 1, the abnormality determination device 1 includes a sensor group 11, a data acquisition unit 12, a storage unit 13, a calculation unit 14, and a display unit 15.
  • the sensor group 11 is composed of a plurality of sensors provided in the blast furnace, and outputs the detected sensor values to the data collection unit 12.
  • Examples of the sensor group 11 include a sensor group installed around the furnace body of the blast furnace.
  • the data collection unit 12 collects the sensor values detected by the sensor group 11 and stores them in the storage unit 13 as operation data. Further, the data collecting unit 12 calculates an index value based on the sensor value detected by the sensor group 11, and also stores the index value as operation data in the storage unit 13.
  • index value based on the sensor value examples include an index value related to the furnace heat of the blast furnace, an index value related to the ventilation resistance of the blast furnace, and the like. Further, as the above-mentioned "index value related to the furnace heat of the blast furnace", an index value calculated from the calorific value of the furnace body and the combustion heat at the tuyere can be mentioned.
  • the storage unit 13 is composed of a recording medium such as an EPROM (Erasable Programmable ROM), a hard disk drive (Hard Disk Drive: HDD), and a removable medium.
  • a recording medium such as an EPROM (Erasable Programmable ROM), a hard disk drive (Hard Disk Drive: HDD), and a removable medium.
  • removable media include disc recording media such as USB (Universal Serial Bus) memory, CD (Compact Disc), DVD (Digital Versatile Disc), and BD (Blu-ray (registered trademark) Disc).
  • the storage unit 13 stores the stable period model 131 constructed by the learning unit 141 and the operation data (sensor value and index value) collected by the data collection unit 12.
  • the operation data stored in the storage unit 13 includes, for example, data (learning data) used when constructing the stable period model 131 (see FIG. 3) and when performing an abnormality determination using the stable period model 131.
  • data (verification data) used for (see FIG. 4).
  • the stable period model 131 is a model constructed based on a plurality of operation data in the stable period of the blast furnace.
  • the stable period model 131 is constructed by the learning unit 141 based on the operation data collected by the data collecting unit 12. The method of constructing the stable period model 131 will be described later.
  • the arithmetic unit 14 is realized by, for example, a processor including a CPU (Central Processing Unit) and a memory (main storage unit) including a RAM (Random Access Memory) and a ROM (Read Only Memory).
  • a processor including a CPU (Central Processing Unit) and a memory (main storage unit) including a RAM (Random Access Memory) and a ROM (Read Only Memory).
  • the arithmetic unit 14 loads and executes the program in the work area of the main storage unit, and controls each component or the like through the execution of the program to realize a function that meets a predetermined purpose.
  • the calculation unit 14 functions as a learning unit (learning means) 141, a difference calculation unit (difference calculation means) 142, and an abnormality determination unit (abnormality determination means) 143 through the execution of the program.
  • learning unit 141, difference calculation unit 142, and abnormality determination unit 143 are realized by one calculation unit ( ⁇ computer), but a plurality of units are realized.
  • the functions of each unit may be realized by the arithmetic unit ( ⁇ computer).
  • the learning unit 141 uses a plurality of operation data in the stable period of the blast furnace (hereinafter referred to as “stable period data”) to perform learning so that the input value and the output value are the same, thereby performing the stable period model 131. To build. Specifically, the learning unit 141 constructs the stable period model 131 by using an autoencoder, which is a method of deep learning. Then, the learning unit 141 stores the constructed stable period model 131 in the storage unit 13. When constructing the stable period data, it is preferable to use the stable period data for at least the past six months.
  • An autoencoder is one of the mechanisms of a neural network, and is a method for reducing the dimension of input data and extracting features.
  • neurons having a “number of dimensions compressed input data” are provided in the intermediate layer. Further, in order to extract the feature amount of the input data, the number of dimensions of the intermediate layer is made smaller than the number of dimensions of the input layer. Then, the output value of the output layer is set to an output value that can reproduce the input value of the input data.
  • the input data is once embedded (encoded) in a small dimension, and the input data is reconstructed based on the encoded data. That is, by encoding with an autoencoder, data can be expressed with a smaller number of dimensions than originally intended.
  • the description of the configuration of the abnormality determination device 1 will be continued.
  • the difference calculation unit 142 inputs a plurality of operation data (verification data) to be determined to the stable period model 131, and calculates the difference between the input value and the output value at that time. Specifically, the difference calculation unit 142 calculates the integrated value of the difference between each input value and each output value when a plurality of operation data are input to the stable period model 131.
  • the above "integrated value of the difference between each input value and each output value” indicates, for example, the integrated value of the absolute value of the difference between each input value and each output value. Further, the difference calculation unit 142 may calculate, for example, the integrated value of the difference between each input value and each output value for each positive or negative, instead of the integrated value of the absolute value of the difference between each input value and each output value. good.
  • the abnormality determination unit 143 determines the operation abnormality of the blast furnace based on the difference between each input value and each output value when a plurality of operation data to be determined are input to the stable period model 131. Specifically, the abnormality determination unit 143 determines that there is an operation abnormality when the integrated value of the difference between each input value and each output value exceeds a preset threshold value, and when it is less than the threshold value, the operation abnormality Judge as none. In this way, by looking at the integrated value of the difference between each input value and each output value to the stable period model 131, small abnormalities of each item are amplified as stacked values, so it is possible to detect operational abnormalities at an early stage. Can be done.
  • the above threshold value can be calculated in advance empirically and experimentally.
  • the abnormality determination unit 143 determines that there is an operation abnormality when the integrated value of the difference on the positive side exceeds the preset threshold on the positive side. Further, the abnormality determination unit 143 determines that there is an operation abnormality when the integrated value of the difference on the negative side exceeds the preset negative threshold value. Further, the abnormality determination unit 143 determines that there is no operation abnormality when the integrated value of the difference on the positive side is less than the preset threshold value on the positive side.
  • the abnormality determination unit 143 determines that there is no operation abnormality when the integrated value of the difference on the negative side is less than the preset negative threshold value. In this way, the abnormality determination device 1 detects the operation abnormality at an early stage because the abnormality on the positive side and the abnormality on the negative side in each item are amplified as accumulated values by performing the abnormality determination for each positive or negative. be able to.
  • the display unit 15 is realized by a display device such as an LCD display or a CRT display, and displays various information based on a display signal input from the calculation unit 14. Examples of the information displayed on the display unit 15 include a stacked graph (see FIG. 5) showing the integrated value of the difference between each input value and each output value for each positive and negative, as described later.
  • model learning method The learning method of the stable period model (hereinafter, referred to as “model learning method”) executed by the abnormality determination device 1 will be described with reference to FIG.
  • the data selection step (step S1) and the model construction step (step S2) are performed in this order.
  • the construction of the stable period model 131 shown in the figure is performed in advance before the abnormality determination method (see FIG. 4) using the stable period model 131 is carried out.
  • the learning unit 141 selects stable period data, which is a plurality of operation data in the stable period of the blast furnace (step S1).
  • stable period data can be selected based on various indicators. For example, when the operating condition of the blast furnace deteriorates, the air flow rate can generally be changed to a small value (wind reduction). Therefore, it is preferable to select the stable period data based on this air flow rate.
  • the operation data excluding the predetermined time before and after the time when the blower flow rate is equal to or higher than the preset threshold value and the blower flow rate becomes equal to or higher than the threshold value is used as the stable period data. It is preferable to select. This is for the following two reasons. (1) It is not desirable as stable period data because there is a possibility that signs of abnormality may be included for several hours before and after the wind reduction. (2) On the contrary, when the air flow rate exceeds the threshold value, it is assumed that the unsteady state is very strong for several hours when the wind of the blast furnace is increased, which is not desirable as the stable period data.
  • the above "predetermined time” is preferably 8 hours, for example. This is because it takes about 8 hours for the raw material charged from the upper part of the blast furnace to descend to the lower part of the furnace, and if it exceeds 8 hours, it is less likely that a factor of deterioration of the operating condition exists in the furnace. Because.
  • the learning unit 141 inputs the stable period data selected in step S1 to the autoencoder to construct the stable period model 131 (step S2).
  • step S2 as shown in FIG. 2 described above, the data set of the stable period data is set in the input and the output by the number of data N, and the stable period model 131 is constructed.
  • step S2 in order to increase the expressivity of the stable period model 131, it is preferable to use stable period data having a large number of data N and having appropriate variations within the stable range. In this way, by using the autoencoder, it is possible to obtain the features of the stable state (that is, the features that express the normal state with reduced dimensions) of the data group of the input stable period data. can.
  • the abnormality determination method executed by the abnormality determination device 1 will be described with reference to FIG.
  • the data input step (step S11), the difference calculation step (step S12), and the abnormality determination step (steps S13 to S15) are performed in this order.
  • the model learning method (FIG. 3) and the abnormality determination method (FIG. 4) are described separately, but the model learning method may be followed by the abnormality determination method.
  • the difference calculation unit 142 inputs the operation data (verification data) to be determined to the stable period model 131 (step S11).
  • the difference calculation unit 142 calculates the integrated value of the difference (error) between each input value and each output value for each input operation data (step S12).
  • step S12 the integrated value of the absolute value of the difference between each input value and each output value may be calculated, or the integrated value of the difference between each input value and each output value is calculated for each positive or negative. You may.
  • calculating the integrated value of the difference for each positive or negative means adding each value of the item having a positive difference and adding each value of the item having a negative difference. This is because the deviation of the positive side and the negative side differs depending on the item, but the item on the positive side is added on the positive side and the item on the negative side is added on the negative side and evaluated respectively. As a result, when an abnormality occurs and the deviation from the normal state becomes large, the integrated value becomes cumulatively large (or small), so that the occurrence of the abnormality can be quickly grasped.
  • the abnormality determination unit 143 determines whether or not the integrated value of the difference between each input value and each output value exceeds a preset threshold value (step S13).
  • the integrated value of the absolute value of the difference between each input value and each output value is calculated in step S12
  • the integrated value and the threshold value are compared in step S13.
  • the integrated value of the difference between each input value and each output value is calculated for each positive or negative in step S12
  • the integrated value of the positive difference and the positive threshold are calculated.
  • the integrated value of the difference on the negative side and the threshold value on the negative side are compared.
  • step S13 If it is determined in step S13 that the integrated value of the difference exceeds a preset threshold value (Yes in step S13), the abnormality determination unit 143 determines that there is an operation abnormality (step S14), and ends this process. On the other hand, in step S13, when it is determined that the integrated value of the difference is less than the preset threshold value (No in step S13), the abnormality determination unit 143 determines that there is no operation abnormality (step S15), and this process is performed. finish.
  • a stacking graph in which the difference between each input value and each output value is stacked for each positive or negative is created, and a display step of displaying the difference on the display unit 15 is performed. May be good. By performing this display step, it is possible to visualize the degree of abnormality for each operation data (input item).
  • blast furnace operation method In the blast furnace operation method, the operation of the blast furnace is changed based on the determination result of the above-mentioned blast furnace abnormality determination method. This makes it possible to prevent serious abnormalities and troubles in the blast furnace.
  • the detection of the abnormality is earlier than in the case where the abnormality determination is performed using the data having the same characteristics. Can be done.
  • various data having different characteristics, which are expected to be related to the operation abnormality are not divided for each characteristic and a model is constructed for each characteristic, but one model is constructed to determine the abnormality. Therefore, the abnormality determination can be performed more easily.
  • the stable period data of the operation of the blast furnace is input to the autoencoder, a stable period model is constructed, and an abnormality determination is performed using the stable period model.
  • the stable period data the operation data when the air flow rate exceeds 75% of the steady operation was used. At that time, the operation data for 8 hours before and after the air flow rate became 75% or less of the steady operation was excluded.
  • FIG. 5 shows the result of inputting the operation data (verification data) before the occurrence of the operation abnormality to the time when the operation abnormality occurs for this stable period model.
  • the vertical axis is the index abnormal value indicating the abnormality of the index value
  • the horizontal axis is the time.
  • the sensor signals used include the measured values of blast furnace exhaust gas (N 2 , H 2 , CO, CO 2 ), the blast furnace gas utilization rate calculated based on them, and the ventilation resistance values of the blast furnace (ventilation of the entire furnace body). , Each ventilation of the lower part / middle part / upper part of the furnace), each index related to the heat of the furnace (heat blast, heat of tuyere combustion, solution reaction heat, heat of top gas, heat of blast moisture decomposition, heat radiated from the furnace body , The amount of heat generated by combustion of pulverized coal, the amount of heat generated by decomposition of pulverized coal, the amount of heat generated by slag, the amount of heat generated by charged raw materials, the heat generated by hot metal) The amount of PCI blown in), the processing value of the sensor group around the furnace body (average for each shaft pressure height, average for each temperature height of each part of the furnace body), etc. are included.
  • verification data we used operation data for about 1.5 months excluding wind breaks. Then, the verification data is input to the stable period model, the difference between the input value and the output value for each item is obtained, and the value obtained by integrating the difference between the input value and the output value for each positive or negative is divided into positive and negative, and FIG. It is represented by a stacked graph as shown in.
  • the blast furnace abnormality determination method, the stable period model learning method, the blast furnace operation method, and the blast furnace abnormality determination device according to the present invention have been specifically described with reference to the embodiments and examples for carrying out the invention.
  • the gist of the invention is not limited to these statements, and must be broadly interpreted based on the statements of the claims. Needless to say, various changes, modifications, etc. based on these descriptions are also included in the gist of the present invention.

Landscapes

  • Engineering & Computer Science (AREA)
  • Chemical & Material Sciences (AREA)
  • Manufacturing & Machinery (AREA)
  • Materials Engineering (AREA)
  • Metallurgy (AREA)
  • Organic Chemistry (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

This abnormality determination method for a blast furnace includes an abnormality determination step for inputting a plurality of pieces of operation data to be determined to a stabilization period model trained by using a plurality of pieces of operation data in a stabilization period of the blast furnace to make input values and output values equal, and determining an abnormality in an operation for the blast furnace on the basis of the difference between the input values and the output values at that time.

Description

高炉の異常判定方法、安定期モデルの学習方法、高炉の操業方法および高炉の異常判定装置Blast furnace abnormality judgment method, stable period model learning method, blast furnace operation method and blast furnace abnormality judgment device
　本発明は、高炉の異常判定方法、安定期モデルの学習方法、高炉の操業方法および高炉の異常判定装置に関する。 The present invention relates to a blast furnace abnormality determination method, a stable period model learning method, a blast furnace operation method, and a blast furnace abnormality determination device.
　プラントでは、操業時に様々な異常が発生する可能性がある。そのため、重大な異常やトラブルを未然に防止するためには、操業時の異常を早期に発見し、適切に対処することが求められる。例えば高炉では、装入した原料の性状の変化や分布の変化、羽口から吹き込む微粉炭の未燃等が原因となり、大きなトラブルが発生する場合がある。 In the plant, various abnormalities may occur during operation. Therefore, in order to prevent serious abnormalities and troubles, it is necessary to detect abnormalities during operation at an early stage and take appropriate measures. For example, in a blast furnace, major troubles may occur due to changes in the properties and distribution of the charged raw materials, unburned pulverized coal blown from the tuyere, and the like.
　プラントの操業異常およびトラブルの予兆は、当該プラントに設置されたセンサで検出されたセンサ値、当該センサ値に基づいて算出された指標値、プラントの画像データ等に含まれていることが多い。従来は、例えば単純に、センサ値ごとまたは指標値ごとに上下限値（閾値）を設定し、その上下限値の範囲を超えた場合に異常と判断していた。しかしながら、この手法では、異常の初期状態（本格的なトラブル前）における操業状態の小さな変化を捉えることは困難である。そこで、例えば特許文献１では、Ｑ統計量を用いて操業異常の予兆を把握する技術が提案されている。 Signs of plant operation abnormalities and troubles are often included in sensor values detected by sensors installed in the plant, index values calculated based on the sensor values, plant image data, and the like. Conventionally, for example, an upper / lower limit value (threshold value) is simply set for each sensor value or index value, and when the range of the upper / lower limit value is exceeded, an abnormality is determined. However, with this method, it is difficult to capture small changes in the operating state in the initial state of abnormality (before full-scale trouble). Therefore, for example, Patent Document 1 proposes a technique for grasping a sign of an operation abnormality by using a Q statistic.
特開２０１７－１２８８０５号公報JP-A-2017-128805
　しかしながら、特許文献１のようにＱ統計量を用いた手法では、例えば正常時の同期性が主成分値に現れるデータ群特性(同文献では、「高炉シャフト圧力の正常時の同期挙動特性」)がある。そのため、特性の異なる様々なデータを同時に用いて異常判定を行うことは困難であった。 However, in the method using the Q statistic as in Patent Document 1, for example, the data group characteristic in which the synchronization in the normal state appears in the principal component value (in the same document, the “synchronous behavior characteristic in the normal state of the blast furnace shaft pressure”). There is. Therefore, it has been difficult to determine an abnormality by using various data having different characteristics at the same time.
　本発明は、上記に鑑みてなされたものであって、特性の異なる様々なデータを同時に用いて、高炉の操業異常の検知を早期に行うことができる高炉の異常判定方法、安定期モデルの学習方法、高炉の操業方法および高炉の異常判定装置を提供することを目的とする。 The present invention has been made in view of the above, and learning of a blast furnace abnormality determination method and a stable period model capable of detecting an operation abnormality of a blast furnace at an early stage by simultaneously using various data having different characteristics. It is an object of the present invention to provide a method, a method of operating a blast furnace, and an abnormality determination device for a blast furnace.
　上述した課題を解決し、目的を達成するために、本発明に係る高炉の異常判定方法は、高炉の安定期における複数の操業データを用いて、入力値と出力値とが同じになるように学習された安定期モデルに対して、判定対象となる複数の操業データを入力し、その際の入力値と出力値との差に基づいて、前記高炉の操業異常を判定する異常判定ステップを含む。 In order to solve the above-mentioned problems and achieve the object, the blast furnace abnormality determination method according to the present invention uses a plurality of operation data in the stable period of the blast furnace so that the input value and the output value are the same. A plurality of operation data to be determined are input to the learned stable period model, and an abnormality determination step of determining an operation abnormality of the blast furnace based on the difference between the input value and the output value at that time is included. ..
　また、本発明に係る高炉の異常判定方法は、上記発明において、前記異常判定ステップは、前記安定期モデルに対して判定対象となる複数の操業データを入力した際の、各入力値と各出力値との差の積算値を算出し、前記差の積算値が、予め設定した閾値を超えた場合に、操業異常ありと判定する。 Further, in the method for determining an abnormality of a blast furnace according to the present invention, in the above invention, the abnormality determination step is each input value and each output when a plurality of operation data to be determined are input to the stable period model. The integrated value of the difference from the value is calculated, and when the integrated value of the difference exceeds a preset threshold value, it is determined that there is an operation abnormality.
　また、本発明に係る高炉の異常判定方法は、上記発明において、前記異常判定ステップが、前記安定期モデルに対して判定対象となる複数の操業データを入力した際の、各入力値と各出力値との差の積算値を、正負ごとに算出し、正側の差の積算値が、予め設定した正側の閾値を超えた場合、あるいは、負側の差の積算値が、予め設定した負側の閾値を超えた場合に、操業異常ありと判定する。 Further, in the method for determining an abnormality of a blast furnace according to the present invention, in the above invention, each input value and each output when the abnormality determination step inputs a plurality of operation data to be determined for the stable period model. The integrated value of the difference from the value is calculated for each positive and negative, and when the integrated value of the positive side difference exceeds the preset positive side threshold value, or the integrated value of the negative side difference is preset. When the threshold on the negative side is exceeded, it is determined that there is an operation abnormality.
　また、本発明に係る高炉の異常判定方法は、上記発明において、前記異常判定ステップの後に、前記安定期モデルに対して判定対象となる複数の操業データを入力した際の、各入力値と各出力値との差の正負ごとの積算値を、積み重ねグラフで表示する表示ステップを更に含む。 Further, in the method for determining an abnormality of a blast furnace according to the present invention, in the above invention, each input value and each input value when a plurality of operation data to be determined are input to the stable period model after the abnormality determination step. It further includes a display step of displaying the integrated value of the difference from the output value for each positive or negative in a stacked graph.
　上述した課題を解決し、目的を達成するために、本発明に係る安定期モデルの学習方法は、高炉の安定期における複数の操業データをオートエンコーダに入力することにより、入力値と出力値とが同じになるように学習させた安定期モデルを構築する学習ステップを含む。 In order to solve the above-mentioned problems and achieve the object, the learning method of the stable period model according to the present invention includes input values and output values by inputting a plurality of operation data in the stable period of the blast furnace to the autoencoder. Includes learning steps to build a stable model trained to be the same.
　また、本発明に係る安定期モデルの学習方法は、上記発明において、前記学習ステップが、送風流量を基準として、前記高炉の安定期における複数の操業データを選択し、前記安定期モデルを構築する。 Further, in the learning method of the stable period model according to the present invention, in the above invention, the learning step selects a plurality of operation data in the stable period of the blast furnace based on the air flow rate, and constructs the stable period model. ..
　また、本発明に係る安定期モデルの学習方法は、上記発明において、前記学習ステップが、前記高炉の複数の操業データのうち、前記送風流量が予め設定した閾値以上であって、かつ前記送風流量が前記閾値以上となった時点の前後の所定時間を除外した操業データを選択し、前記安定期モデルを構築する。 Further, in the learning method of the stable period model according to the present invention, in the above invention, the learning step is such that the blast flow rate is equal to or more than a preset threshold value among a plurality of operation data of the blast furnace, and the blast flow rate is the same. The operation data excluding the predetermined time before and after the time when is equal to or more than the threshold value is selected, and the stable period model is constructed.
　上述した課題を解決し、目的を達成するために、本発明に係る高炉の操業方法は、上記の高炉の異常判定方法の判定結果に基づいて、高炉の操業を変更する。 In order to solve the above-mentioned problems and achieve the object, the blast furnace operation method according to the present invention changes the operation of the blast furnace based on the determination result of the above-mentioned blast furnace abnormality determination method.
　上述した課題を解決し、目的を達成するために、本発明に係る高炉の異常判定装置は、高炉の安定期における複数の操業データを用いて、入力値と出力値とが同じになるように学習された安定期モデルに対して、判定対象となる複数の操業データを入力し、その際の入力値と出力値との差に基づいて、前記高炉の操業異常を判定する異常判定手段を備える。 In order to solve the above-mentioned problems and achieve the object, the blast furnace abnormality determination device according to the present invention uses a plurality of operation data in the stable period of the blast furnace so that the input value and the output value are the same. A plurality of operation data to be determined are input to the learned stable period model, and an abnormality determination means for determining an operation abnormality of the blast furnace is provided based on the difference between the input value and the output value at that time. ..
　本発明によれば、入力値と出力値とが同じになるように学習された安定期モデルを用いて異常判定を行うことにより、特性の異なる様々なデータを同時に用いて、高炉の操業異常の検知を早期に行うことができる。 According to the present invention, by performing anomaly determination using a stable period model trained so that the input value and the output value are the same, various data having different characteristics can be used simultaneously to determine the operation abnormality of the blast furnace. Detection can be performed early.
図１は、本発明の実施形態に係る高炉の異常判定装置の概略的な構成を示す図である。FIG. 1 is a diagram showing a schematic configuration of an abnormality determination device for a blast furnace according to an embodiment of the present invention. 図２は、本発明の実施形態に係る安定期モデルの学習方法で用いるオートエンコーダの概要を示す図である。FIG. 2 is a diagram showing an outline of an autoencoder used in the learning method of the stable period model according to the embodiment of the present invention. 図３は、本発明の実施形態に係る安定期モデルの学習方法の流れを示すフローチャートである。FIG. 3 is a flowchart showing the flow of the learning method of the stable period model according to the embodiment of the present invention. 図４は、本発明の実施形態に係る異常判定方法の流れを示すフローチャートである。FIG. 4 is a flowchart showing the flow of the abnormality determination method according to the embodiment of the present invention. 図５は、本発明に係る高炉の異常判定方法の実施例であり、所定期間の検証データを用いて異常判定を行った一例を示す図である。FIG. 5 is an example of an abnormality determination method for a blast furnace according to the present invention, and is a diagram showing an example in which an abnormality determination is performed using verification data for a predetermined period.
　本発明の実施形態に係る高炉の異常判定方法、安定期モデルの学習方法、高炉の操業方法および高炉の異常判定装置について、図面を参照しながら説明する。以下では、高炉の異常判定装置、安定期モデルの学習方法、高炉の異常判定方法、高炉の操業方法の順に説明を行う。なお、本発明は以下で説明する実施形態に限定されるものではない。 The blast furnace abnormality determination method, the learning method of the stable period model, the blast furnace operation method, and the blast furnace abnormality determination device according to the embodiment of the present invention will be described with reference to the drawings. In the following, the blast furnace abnormality determination device, the learning method of the stable period model, the blast furnace abnormality determination method, and the blast furnace operation method will be described in this order. The present invention is not limited to the embodiments described below.
（異常判定装置）
　本発明の実施形態に係る高炉の異常判定装置の構成について、図１を参照しながら説明する。異常判定装置１は、高炉等のプラントの異常を判定するためのものである。異常判定装置１は、図１に示すように、センサ群１１と、データ収集部１２と、記憶部１３と、演算部１４と、表示部１５と、を備えている。 (Abnormality judgment device)
The configuration of the abnormality determination device for the blast furnace according to the embodiment of the present invention will be described with reference to FIG. The abnormality determination device 1 is for determining an abnormality in a plant such as a blast furnace. As shown in FIG. 1, the abnormality determination device 1 includes a sensor group 11, a data acquisition unit 12, a storage unit 13, a calculation unit 14, and a display unit 15.
　センサ群１１は、高炉に設けられた複数のセンサからなり、検出したセンサ値をデータ収集部１２に出力する。センサ群１１としては、例えば高炉の炉体周りに設置されたセンサ群等が挙げられる。 The sensor group 11 is composed of a plurality of sensors provided in the blast furnace, and outputs the detected sensor values to the data collection unit 12. Examples of the sensor group 11 include a sensor group installed around the furnace body of the blast furnace.
　データ収集部１２は、センサ群１１が検出したセンサ値を収集し、操業データとして記憶部１３に蓄積する。また、データ収集部１２は、センサ群１１が検出したセンサ値に基づいて指標値を算出し、当該指標値についても、操業データとして記憶部１３に蓄積する。 The data collection unit 12 collects the sensor values detected by the sensor group 11 and stores them in the storage unit 13 as operation data. Further, the data collecting unit 12 calculates an index value based on the sensor value detected by the sensor group 11, and also stores the index value as operation data in the storage unit 13.
　上記の「センサ値に基づく指標値（以下、単に「指標値」という）」としては、高炉の炉熱に関する指標値、高炉の通気抵抗に関する指標値等が挙げられる。また、上記の「高炉の炉熱に関する指標値」としては、炉体の熱量や羽口先における燃焼熱から算出される指標値等が挙げられる。 Examples of the above-mentioned "index value based on the sensor value (hereinafter, simply referred to as" index value ")" include an index value related to the furnace heat of the blast furnace, an index value related to the ventilation resistance of the blast furnace, and the like. Further, as the above-mentioned "index value related to the furnace heat of the blast furnace", an index value calculated from the calorific value of the furnace body and the combustion heat at the tuyere can be mentioned.
　記憶部１３は、例えばＥＰＲＯＭ（Erasable　Programmable　ROM）、ハードディスクドライブ（Hard　Disk　Drive：ＨＤＤ）およびリムーバブルメディア等の記録媒体から構成される。リムーバブルメディアとしては、例えばＵＳＢ（Universal　Serial　Bus）メモリ、ＣＤ（Compact　Disc）、ＤＶＤ（Digital　Versatile　Disc）、ＢＤ（Blu-ray（登録商標）　Disc）のようなディスク記録媒体が挙げられる。 The storage unit 13 is composed of a recording medium such as an EPROM (Erasable Programmable ROM), a hard disk drive (Hard Disk Drive: HDD), and a removable medium. Examples of removable media include disc recording media such as USB (Universal Serial Bus) memory, CD (Compact Disc), DVD (Digital Versatile Disc), and BD (Blu-ray (registered trademark) Disc).
　記憶部１３には、学習部１４１によって構築された安定期モデル１３１と、データ収集部１２によって収集された操業データ（センサ値および指標値）とが格納されている。なお、記憶部１３に格納される操業データには、例えば安定期モデル１３１を構築する際（図３参照）に用いられるデータ（学習データ）と、安定期モデル１３１を用いて異常判定を行う際（図４参照）に用いられるデータ（検証データ）とがある。 The storage unit 13 stores the stable period model 131 constructed by the learning unit 141 and the operation data (sensor value and index value) collected by the data collection unit 12. The operation data stored in the storage unit 13 includes, for example, data (learning data) used when constructing the stable period model 131 (see FIG. 3) and when performing an abnormality determination using the stable period model 131. There is data (verification data) used for (see FIG. 4).
　安定期モデル１３１は、高炉の安定期における複数の操業データに基づいて構築されたモデルである。この安定期モデル１３１は、学習部１４１によって、データ収集部１２によって収集された操業データに基づいて構築される。安定期モデル１３１の構築方法については後記する。 The stable period model 131 is a model constructed based on a plurality of operation data in the stable period of the blast furnace. The stable period model 131 is constructed by the learning unit 141 based on the operation data collected by the data collecting unit 12. The method of constructing the stable period model 131 will be described later.
　演算部１４は、例えばＣＰＵ（Central　Processing　Unit）等からなるプロセッサと、ＲＡＭ（Random　Access　Memory）やＲＯＭ（Read　Only　Memory）等からなるメモリ（主記憶部）と、によって実現される。 The arithmetic unit 14 is realized by, for example, a processor including a CPU (Central Processing Unit) and a memory (main storage unit) including a RAM (Random Access Memory) and a ROM (Read Only Memory).
　演算部１４は、プログラムを主記憶部の作業領域にロードして実行し、プログラムの実行を通じて各構成部等を制御することにより、所定の目的に合致した機能を実現する。演算部１４は、具体的にはプログラムの実行を通じて、学習部（学習手段）１４１、差分算出部（差分算出手段）１４２および異常判定部（異常判定手段）１４３として機能する。なお、本実施形態では、図１に示すように、一つの演算部（≒コンピュータ）によって各部（学習部１４１、差分算出部１４２および異常判定部１４３）の機能を実現しているが、複数の演算部（≒コンピュータ）により各部の機能をそれぞれ実現してもよい。 The arithmetic unit 14 loads and executes the program in the work area of the main storage unit, and controls each component or the like through the execution of the program to realize a function that meets a predetermined purpose. Specifically, the calculation unit 14 functions as a learning unit (learning means) 141, a difference calculation unit (difference calculation means) 142, and an abnormality determination unit (abnormality determination means) 143 through the execution of the program. In the present embodiment, as shown in FIG. 1, the functions of each unit (learning unit 141, difference calculation unit 142, and abnormality determination unit 143) are realized by one calculation unit (≈ computer), but a plurality of units are realized. The functions of each unit may be realized by the arithmetic unit (≈ computer).
　学習部１４１は、高炉の安定期における複数の操業データ（以下、「安定期データ」という）を用いて、入力値と出力値とが同じになるように学習を行うことにより、安定期モデル１３１を構築する。学習部１４１は、具体的には、深層学習の一手法であるオートエンコーダを用いて、安定期モデル１３１を構築する。そして、学習部１４１は、構築した安定期モデル１３１を記憶部１３に格納する。なお、安定期データを構築する際は、少なくとも過去半年分の安定期データを用いることが好ましい。 The learning unit 141 uses a plurality of operation data in the stable period of the blast furnace (hereinafter referred to as “stable period data”) to perform learning so that the input value and the output value are the same, thereby performing the stable period model 131. To build. Specifically, the learning unit 141 constructs the stable period model 131 by using an autoencoder, which is a method of deep learning. Then, the learning unit 141 stores the constructed stable period model 131 in the storage unit 13. When constructing the stable period data, it is preferable to use the stable period data for at least the past six months.
　学習部１４１が用いるオートエンコーダの概要について、図２を参照しながら説明する。オートエンコーダは、ニューラルネットワークの仕組みの一つであり、入力データの次元を削減し、特徴量を抽出するための手法である。オートエンコーダでは、図２に示すように、「入力データを圧縮した次元数」のニューロンを中間層に設ける。また、入力データの特徴量を抽出するために、中間層の次元数を入力層の次元数よりも小さくする。そして、出力層の出力値を、入力データの入力値を再現できるような出力値とする。 The outline of the autoencoder used by the learning unit 141 will be described with reference to FIG. An autoencoder is one of the mechanisms of a neural network, and is a method for reducing the dimension of input data and extracting features. In the autoencoder, as shown in FIG. 2, neurons having a “number of dimensions compressed input data” are provided in the intermediate layer. Further, in order to extract the feature amount of the input data, the number of dimensions of the intermediate layer is made smaller than the number of dimensions of the input layer. Then, the output value of the output layer is set to an output value that can reproduce the input value of the input data.
　このような構成を有するオートエンコーダでは、入力データを一度小さい次元に埋め込み（エンコードし）、エンコードしたデータに基づいて入力データの再構築を行う。すなわち、オートエンコーダによってエンコードすることにより、本来よりも小さな次元数でデータを表現できることになる。図１に戻って異常判定装置１の構成の説明を続ける。 In an autoencoder having such a configuration, the input data is once embedded (encoded) in a small dimension, and the input data is reconstructed based on the encoded data. That is, by encoding with an autoencoder, data can be expressed with a smaller number of dimensions than originally intended. Returning to FIG. 1, the description of the configuration of the abnormality determination device 1 will be continued.
　差分算出部１４２は、安定期モデル１３１に対して、判定対象となる複数の操業データ（検証データ）を入力し、その際の入力値と出力値との差を算出する。差分算出部１４２は、具体的には、安定期モデル１３１に対して複数の操業データを入力した際の、各入力値と各出力値との差の積算値を算出する。 The difference calculation unit 142 inputs a plurality of operation data (verification data) to be determined to the stable period model 131, and calculates the difference between the input value and the output value at that time. Specifically, the difference calculation unit 142 calculates the integrated value of the difference between each input value and each output value when a plurality of operation data are input to the stable period model 131.
　上記の「各入力値と各出力値との差の積算値」とは、例えば各入力値と各出力値との差の絶対値を積算した値のことを示している。また、差分算出部１４２は、各入力値と各出力値との差の絶対値の積算値ではなく、例えば各入力値と各出力値との差の積算値を、正負ごとに算出してもよい。 The above "integrated value of the difference between each input value and each output value" indicates, for example, the integrated value of the absolute value of the difference between each input value and each output value. Further, the difference calculation unit 142 may calculate, for example, the integrated value of the difference between each input value and each output value for each positive or negative, instead of the integrated value of the absolute value of the difference between each input value and each output value. good.
　異常判定部１４３は、安定期モデル１３１に対して判定対象となる複数の操業データを入力した際の、各入力値と各出力値との差に基づいて、高炉の操業異常を判定する。異常判定部１４３は、具体的には、各入力値と各出力値との差の積算値が、予め設定した閾値を超えた場合に操業異常ありと判定し、閾値未満である場合に操業異常なしと判定する。このように、安定期モデル１３１への各入力値と各出力値との差の積算値を見ることにより、各項目の小さな異常が積み重ね値として増幅されるため、操業異常を早期に検知することができる。なお、上記の閾値は、予め経験的および実験的に算出することができる。 The abnormality determination unit 143 determines the operation abnormality of the blast furnace based on the difference between each input value and each output value when a plurality of operation data to be determined are input to the stable period model 131. Specifically, the abnormality determination unit 143 determines that there is an operation abnormality when the integrated value of the difference between each input value and each output value exceeds a preset threshold value, and when it is less than the threshold value, the operation abnormality Judge as none. In this way, by looking at the integrated value of the difference between each input value and each output value to the stable period model 131, small abnormalities of each item are amplified as stacked values, so it is possible to detect operational abnormalities at an early stage. Can be done. The above threshold value can be calculated in advance empirically and experimentally.
　ここで、上記の差分算出部１４２において、各入力値と各出力値との差の積算値を、正負ごとに算出した場合、次のような処理を行う。この場合、異常判定部１４３は、正側の差の積算値が、予め設定した正側の閾値を超えた場合に、操業異常ありと判定する。また、異常判定部１４３は、負側の差の積算値が、予め設定した負側の閾値を超えた場合に、操業異常ありと判定する。また、異常判定部１４３は、正側の差の積算値が、予め設定した正側の閾値未満である場合に、操業異常なしと判定する。また、異常判定部１４３は、負側の差の積算値が、予め設定した負側の閾値未満である場合に、操業異常なしと判定する。このように、異常判定装置１では、正負ごとに異常判定を行うことにより、各項目における正側の異常と負側の異常とがそれぞれ積み重ね値として増幅されるため、操業異常を早期に検知することができる。 Here, when the integrated value of the difference between each input value and each output value is calculated for each positive or negative in the above difference calculation unit 142, the following processing is performed. In this case, the abnormality determination unit 143 determines that there is an operation abnormality when the integrated value of the difference on the positive side exceeds the preset threshold on the positive side. Further, the abnormality determination unit 143 determines that there is an operation abnormality when the integrated value of the difference on the negative side exceeds the preset negative threshold value. Further, the abnormality determination unit 143 determines that there is no operation abnormality when the integrated value of the difference on the positive side is less than the preset threshold value on the positive side. Further, the abnormality determination unit 143 determines that there is no operation abnormality when the integrated value of the difference on the negative side is less than the preset negative threshold value. In this way, the abnormality determination device 1 detects the operation abnormality at an early stage because the abnormality on the positive side and the abnormality on the negative side in each item are amplified as accumulated values by performing the abnormality determination for each positive or negative. be able to.
　表示部１５は、例えばＬＣＤディスプレイ、ＣＲＴディスプレイ等の表示装置によって実現され、演算部１４から入力される表示信号に基づいて各種情報を表示する。表示部１５で表示する情報としては、例えば後記するように、各入力値と各出力値との差の正負ごとの積算値を示す積み重ねグラフ（図５参照）等が挙げられる。 The display unit 15 is realized by a display device such as an LCD display or a CRT display, and displays various information based on a display signal input from the calculation unit 14. Examples of the information displayed on the display unit 15 include a stacked graph (see FIG. 5) showing the integrated value of the difference between each input value and each output value for each positive and negative, as described later.
（安定期モデルの学習方法）
　異常判定装置１が実行する安定期モデルの学習方法（以下、「モデル学習方法」という）について、図３を参照しながら説明する。モデル学習方法では、データ選択工程（ステップＳ１）と、モデル構築工程（ステップＳ２）と、をこの順で行う。なお、同図に示した安定期モデル１３１の構築は、安定期モデル１３１を用いた異常判定方法（図４参照）を実施する前に、予め行っておく。 (Learning method of stable period model)
The learning method of the stable period model (hereinafter, referred to as “model learning method”) executed by the abnormality determination device 1 will be described with reference to FIG. In the model learning method, the data selection step (step S1) and the model construction step (step S2) are performed in this order. The construction of the stable period model 131 shown in the figure is performed in advance before the abnormality determination method (see FIG. 4) using the stable period model 131 is carried out.
　データ選択工程では、学習部１４１が、高炉の安定期における複数の操業データである安定期データを選択する（ステップＳ１）。ステップＳ１では、様々な指標を基準として安定期データを選択することができるが、例えば高炉の操業状態が悪化した場合は、一般的に送風流量を小さな値に変更する（減風する）ことが行われるため、この送風流量を基準として安定期データを選択することが好ましい。 In the data selection step, the learning unit 141 selects stable period data, which is a plurality of operation data in the stable period of the blast furnace (step S1). In step S1, stable period data can be selected based on various indicators. For example, when the operating condition of the blast furnace deteriorates, the air flow rate can generally be changed to a small value (wind reduction). Therefore, it is preferable to select the stable period data based on this air flow rate.
　この場合、高炉の複数の操業データのうち、送風流量が予め設定した閾値以上であって、かつ送風流量が閾値以上となった時点の前後の所定時間を除外した操業データを、安定期データとして選択することが好ましい。これは、以下の二つの理由からである。
（１）減風前後の数時間は異常の予兆が含まれている可能性があるため、安定期データとしては望ましくない。
（２）逆に、送風流量が閾値を超えた場合、高炉の増風時の数時間は非定常性が非常に強いことが想定されるため、安定期データとしては望ましくない。 In this case, among the plurality of operation data of the blast furnace, the operation data excluding the predetermined time before and after the time when the blower flow rate is equal to or higher than the preset threshold value and the blower flow rate becomes equal to or higher than the threshold value is used as the stable period data. It is preferable to select. This is for the following two reasons.
(1) It is not desirable as stable period data because there is a possibility that signs of abnormality may be included for several hours before and after the wind reduction.
(2) On the contrary, when the air flow rate exceeds the threshold value, it is assumed that the unsteady state is very strong for several hours when the wind of the blast furnace is increased, which is not desirable as the stable period data.
　上記の「所定時間」は、例えば８時間とすることが好ましい。これは、高炉上部から装入された原料が炉下部まで降下するまでの時間が８時間程度であり、８時間を超えると、操業状態の悪化の要因が炉内に存在する可能性も低くなるためである。 The above "predetermined time" is preferably 8 hours, for example. This is because it takes about 8 hours for the raw material charged from the upper part of the blast furnace to descend to the lower part of the furnace, and if it exceeds 8 hours, it is less likely that a factor of deterioration of the operating condition exists in the furnace. Because.
　続いて、モデル構築工程では、学習部１４１が、ステップＳ１で選択した安定期データをオートエンコーダに入力し、安定期モデル１３１を構築する（ステップＳ２）。ステップＳ２では、前記した図２に示すように、安定期データのデータセットを、データ数Ｎ分だけ入力と出力にセットし、安定期モデル１３１を構築する。なお、ステップＳ２では、安定期モデル１３１の表現度を上げるために、データ数Ｎが多く、かつ安定範囲内で適度なばらつきのある安定期データを用いることが好ましい。このように、オートエンコーダを用いることにより、入力した安定期データのデータ群の、安定状態の特徴量（すなわち次元が削減された、正常状態を表現する特徴を数値化したもの）を求めることができる。 Subsequently, in the model construction step, the learning unit 141 inputs the stable period data selected in step S1 to the autoencoder to construct the stable period model 131 (step S2). In step S2, as shown in FIG. 2 described above, the data set of the stable period data is set in the input and the output by the number of data N, and the stable period model 131 is constructed. In step S2, in order to increase the expressivity of the stable period model 131, it is preferable to use stable period data having a large number of data N and having appropriate variations within the stable range. In this way, by using the autoencoder, it is possible to obtain the features of the stable state (that is, the features that express the normal state with reduced dimensions) of the data group of the input stable period data. can.
（異常判定方法）
　異常判定装置１が実行する異常判定方法について、図４を参照しながら説明する。異常判定方法では、データ入力工程（ステップＳ１１）と、差分算出工程（ステップＳ１２）と、異常判定工程（ステップＳ１３～Ｓ１５）と、をこの順で行う。なお、本実施形態では、モデル学習方法（図３）と異常判定方法（図４）とを分けて説明しているが、モデル学習方法の後に異常判定方法を続けて実施してもよい。 (Abnormality judgment method)
The abnormality determination method executed by the abnormality determination device 1 will be described with reference to FIG. In the abnormality determination method, the data input step (step S11), the difference calculation step (step S12), and the abnormality determination step (steps S13 to S15) are performed in this order. In the present embodiment, the model learning method (FIG. 3) and the abnormality determination method (FIG. 4) are described separately, but the model learning method may be followed by the abnormality determination method.
　データ入力工程では、差分算出部１４２が、安定期モデル１３１に対して判定対象の操業データ（検証データ）を入力する（ステップＳ１１）。 In the data input process, the difference calculation unit 142 inputs the operation data (verification data) to be determined to the stable period model 131 (step S11).
　続いて、差分算出工程では、差分算出部１４２が、入力した操業データごとに、各入力値と各出力値との差（誤差）の積算値を算出する（ステップＳ１２）。なお、ステップＳ１２では、各入力値と各出力値との差の絶対値の積算値を算出してもよく、あるいは各入力値と各出力値との差の積算値を、正負ごとに算出してもよい。 Subsequently, in the difference calculation step, the difference calculation unit 142 calculates the integrated value of the difference (error) between each input value and each output value for each input operation data (step S12). In step S12, the integrated value of the absolute value of the difference between each input value and each output value may be calculated, or the integrated value of the difference between each input value and each output value is calculated for each positive or negative. You may.
　ここで、「差の積算値を正負ごとに算出する」とは、差が正の項目の各値を加算し、かつ差が負の項目の各値を加算することを意味する。これは、項目によって正側、負側のずれ方が異なるが、正側の項目は正側で加算し、負側の項目は負側で加算し、それぞれ評価するものである。これにより、異常発生時に、正常な状態からのずれが大きくなっていく場合に、積算値が累積的に大きく（または小さく）なるため、異常発生を速やかに把握することが可能となる。 Here, "calculating the integrated value of the difference for each positive or negative" means adding each value of the item having a positive difference and adding each value of the item having a negative difference. This is because the deviation of the positive side and the negative side differs depending on the item, but the item on the positive side is added on the positive side and the item on the negative side is added on the negative side and evaluated respectively. As a result, when an abnormality occurs and the deviation from the normal state becomes large, the integrated value becomes cumulatively large (or small), so that the occurrence of the abnormality can be quickly grasped.
　続いて、異常判定工程では、異常判定部１４３が、各入力値と各出力値との差の積算値が、予め設定した閾値を超えるか否かを判定する（ステップＳ１３）。ここで、上記のステップＳ１２において、各入力値と各出力値との差の絶対値の積算値を算出している場合、ステップＳ１３では、当該積算値と閾値とを比較する。一方、上記のステップＳ１２において、各入力値と各出力値との差の積算値を、正負ごとに算出している場合、ステップＳ１３では、正側の差の積算値と正側の閾値とを比較するとともに、負側の差の積算値と負側の閾値とを比較する。 Subsequently, in the abnormality determination step, the abnormality determination unit 143 determines whether or not the integrated value of the difference between each input value and each output value exceeds a preset threshold value (step S13). Here, when the integrated value of the absolute value of the difference between each input value and each output value is calculated in step S12, the integrated value and the threshold value are compared in step S13. On the other hand, when the integrated value of the difference between each input value and each output value is calculated for each positive or negative in step S12, in step S13, the integrated value of the positive difference and the positive threshold are calculated. Along with the comparison, the integrated value of the difference on the negative side and the threshold value on the negative side are compared.
　ステップＳ１３において、差の積算値が、予め設定した閾値を超えると判定した場合（ステップＳ１３でＹｅｓ）、異常判定部１４３は、操業異常ありと判定し（ステップＳ１４）、本処理を終了する。一方、ステップＳ１３において、差の積算値が、予め設定した閾値未満であると判定した場合（ステップＳ１３でＮｏ）、異常判定部１４３は、操業異常なしと判定し（ステップＳ１５）、本処理を終了する。 If it is determined in step S13 that the integrated value of the difference exceeds a preset threshold value (Yes in step S13), the abnormality determination unit 143 determines that there is an operation abnormality (step S14), and ends this process. On the other hand, in step S13, when it is determined that the integrated value of the difference is less than the preset threshold value (No in step S13), the abnormality determination unit 143 determines that there is no operation abnormality (step S15), and this process is performed. finish.
　なお、異常判定方法では、上記のステップＳ１１～Ｓ１５に加えて、各入力値と各出力値との差を正負ごとに積み重ねた積み重ねグラフを作成し、表示部１５に表示する表示ステップを行ってもよい。この表示ステップを行うことにより、操業データ（入力項目）ごとの異常度を可視化することができる。 In the abnormality determination method, in addition to the above steps S11 to S15, a stacking graph in which the difference between each input value and each output value is stacked for each positive or negative is created, and a display step of displaying the difference on the display unit 15 is performed. May be good. By performing this display step, it is possible to visualize the degree of abnormality for each operation data (input item).
（高炉の操業方法）
　高炉の操業方法では、上記の高炉の異常判定方法の判定結果に基づいて、高炉の操業を変更する。これにより、高炉における重大な異常やトラブルを未然に防止することができる。 (Blast furnace operation method)
In the blast furnace operation method, the operation of the blast furnace is changed based on the determination result of the above-mentioned blast furnace abnormality determination method. This makes it possible to prevent serious abnormalities and troubles in the blast furnace.
　以上説明した、本実施形態に係る高炉の異常判定方法、安定期モデルの学習方法、高炉の操業方法および高炉の異常判定装置１によれば、入力値と出力値とが同じになるように学習された安定期モデル１３１を用いて異常判定を行う。これにより、特性の異なる様々なデータを同時に用いて、高炉の操業異常の検知を早期に行うことができる。 According to the above-described blast furnace abnormality determination method, stable period model learning method, blast furnace operation method, and blast furnace abnormality determination device 1 according to the present embodiment, learning is performed so that the input value and the output value are the same. Anomaly determination is performed using the stable period model 131. As a result, it is possible to detect an operation abnormality of the blast furnace at an early stage by using various data having different characteristics at the same time.
　また、本実施形態によれば、安定期モデル１３１への各入力値と各出力値との差の積算値を見ることにより、各項目の小さな異常が積み重ね値として増幅されるため、操業異常を早期に検知することができる。 Further, according to the present embodiment, by looking at the integrated value of the difference between each input value and each output value in the stable period model 131, small abnormalities of each item are amplified as stacked values, so that an operation abnormality is caused. It can be detected early.
　また、本実施形態によれば、特性の異なる様々なデータを同時に用いて異常判定を行うことにより、同じ特性のデータを用いて異常判定を行う場合と比較して、より異常の検知をより早期に行うことができる。 Further, according to the present embodiment, by performing the abnormality determination using various data having different characteristics at the same time, the detection of the abnormality is earlier than in the case where the abnormality determination is performed using the data having the same characteristics. Can be done.
　また、本実施形態では、操業異常に関連することが予想される、特性の異なる様々なデータを、特性ごとに分けてモデルを特性別に構築するのではなく、一つのモデルを構築して異常判定を行うため、異常判定をより簡易に行うことができる。 Further, in the present embodiment, various data having different characteristics, which are expected to be related to the operation abnormality, are not divided for each characteristic and a model is constructed for each characteristic, but one model is constructed to determine the abnormality. Therefore, the abnormality determination can be performed more easily.
（実施例）
　本発明の実施例について、図５を参照しながら説明する。本実施例では、高炉の操業の安定期データをオートエンコーダに入力し、安定期モデルを構築し、当該安定期モデルを用いて異常判定を行った。本実施例では、安定期データとして、送風流量が定常操業の７５％超えであるときの操業データを用いた。その際、送風流量が定常操業の７５％以下になった前後８時間分の操業データは除外した。 (Example)
Examples of the present invention will be described with reference to FIG. In this embodiment, the stable period data of the operation of the blast furnace is input to the autoencoder, a stable period model is constructed, and an abnormality determination is performed using the stable period model. In this example, as the stable period data, the operation data when the air flow rate exceeds 75% of the steady operation was used. At that time, the operation data for 8 hours before and after the air flow rate became 75% or less of the steady operation was excluded.
　安定期データとして、高炉の操業異常に関わる可能性のある３６項目に対して１時間周期で収集したデータを約１．５年間分用いて、安定期モデルを構築した。この安定期モデルに対して、操業異常発生前～操業異常発生時の操業データ（検証データ）を入力した結果を図５に示す。なお、同図の縦軸は指標値の異常を示す指標異常値、横軸は時間である。 As stable period data, a stable period model was constructed using data collected in an hourly cycle for 36 items that may be related to blast furnace operation abnormalities for about 1.5 years. FIG. 5 shows the result of inputting the operation data (verification data) before the occurrence of the operation abnormality to the time when the operation abnormality occurs for this stable period model. In the figure, the vertical axis is the index abnormal value indicating the abnormality of the index value, and the horizontal axis is the time.
　使用したセンサ信号としては、高炉排ガスの各計測値（Ｎ２、Ｈ２、ＣＯ、ＣＯ２）や、それに基づいて計算される高炉ガス利用率、高炉の各通気抵抗値（炉体全体の通気、炉下部／炉中部／炉上部の各通気）、炉熱関係の各指数（送風顕熱量、羽口先燃焼熱量、ソリューション反応熱量、炉頂ガス顕熱量、送風湿分分解熱量、炉体放散熱量、微粉炭燃焼熱量、微粉炭分解熱量、スラグ顕熱量、装入原料顕熱量、溶銑顕熱）、操業操作量に係わる値（送風流量、送風圧力、炉頂圧力、送風湿分、送風温度、ＰＣＩ吹き込み量）、炉体周りセンサ群の加工値（シャフト圧力高さ毎平均、炉体各部温度高さ毎平均）等が含まれている。 The sensor signals used include the measured values of blast furnace exhaust gas (N 2 , H 2 , CO, CO 2 ), the blast furnace gas utilization rate calculated based on them, and the ventilation resistance values of the blast furnace (ventilation of the entire furnace body). , Each ventilation of the lower part / middle part / upper part of the furnace), each index related to the heat of the furnace (heat blast, heat of tuyere combustion, solution reaction heat, heat of top gas, heat of blast moisture decomposition, heat radiated from the furnace body , The amount of heat generated by combustion of pulverized coal, the amount of heat generated by decomposition of pulverized coal, the amount of heat generated by slag, the amount of heat generated by charged raw materials, the heat generated by hot metal) The amount of PCI blown in), the processing value of the sensor group around the furnace body (average for each shaft pressure height, average for each temperature height of each part of the furnace body), etc. are included.
　検証データとして、休風時を除いた約１．５ヶ月の操業データを用いた。そして、検証データを安定期モデルに入力し、その項目ごとの入力値と出力値との差を求め、入力値と出力値との差を正負ごとに積算した値を正負に分けて、図５に示すような積み重ねグラフで表現した。 As verification data, we used operation data for about 1.5 months excluding wind breaks. Then, the verification data is input to the stable period model, the difference between the input value and the output value for each item is obtained, and the value obtained by integrating the difference between the input value and the output value for each positive or negative is divided into positive and negative, and FIG. It is represented by a stacked graph as shown in.
　図５に示すように、オペレータが操業異常を認識して対応を取る約１０日前から異常の予兆が現れ、様々な指標値が異常方向に増大していることがわかる。そして、約６日前に、設定した正側の閾値を超えて、異常と判断されている。この時点でオペレータに提示することにより、事前対処が可能となり、被害を回避または低減できる可能性がより高くなる。また、図５において、項目ごとの入力値と出力値との差を、項目ごとに色を変えて積み重ねグラフで表現することにより、どの項目が異常であるのかについても、オペレータが目視で把握することが可能となる。 As shown in FIG. 5, it can be seen that signs of abnormality appear about 10 days before the operator recognizes the operation abnormality and takes action, and various index values are increasing in the abnormality direction. Then, about 6 days ago, the threshold value on the positive side that was set was exceeded, and it was determined to be abnormal. By presenting it to the operator at this point, it is possible to take proactive measures and it is more likely that damage can be avoided or reduced. Further, in FIG. 5, by expressing the difference between the input value and the output value for each item in a stacked graph by changing the color for each item, the operator can visually grasp which item is abnormal. It becomes possible.
　以上、本発明に係る高炉の異常判定方法、安定期モデルの学習方法、高炉の操業方法および高炉の異常判定装置について、発明を実施するための形態および実施例により具体的に説明したが、本発明の趣旨はこれらの記載に限定されるものではなく、請求の範囲の記載に基づいて広く解釈されなければならない。また、これらの記載に基づいて種々変更、改変等したものも本発明の趣旨に含まれることはいうまでもない。 The blast furnace abnormality determination method, the stable period model learning method, the blast furnace operation method, and the blast furnace abnormality determination device according to the present invention have been specifically described with reference to the embodiments and examples for carrying out the invention. The gist of the invention is not limited to these statements, and must be broadly interpreted based on the statements of the claims. Needless to say, various changes, modifications, etc. based on these descriptions are also included in the gist of the present invention.
　１　異常判定装置
　１１　センサ群
　１２　データ収集部
　１３　記憶部
　１３１　安定期モデル
　１４　演算部
　１４１　学習部
　１４２　差分算出部
　１４３　異常判定部
　１５　表示部 1 Abnormality judgment device 11 Sensor group 12 Data collection unit 13 Storage unit 131 Stable period model 14 Calculation unit 141 Learning unit 142 Difference calculation unit 143 Abnormality determination unit 15 Display unit

Claims (9)

  1. 　高炉の安定期における複数の操業データを用いて、入力値と出力値とが同じになるように学習された安定期モデルに対して、判定対象となる複数の操業データを入力し、その際の入力値と出力値との差に基づいて、前記高炉の操業異常を判定する異常判定ステップを含む高炉の異常判定方法。 Using multiple operation data in the stable period of the blast furnace, multiple operation data to be judged are input to the stable period model trained so that the input value and the output value are the same, and at that time. A method for determining an abnormality in a blast furnace, which includes an abnormality determination step for determining an operation abnormality in the blast furnace based on a difference between an input value and an output value.
  2. 　前記異常判定ステップは、
    　前記安定期モデルに対して判定対象となる複数の操業データを入力した際の、各入力値と各出力値との差の積算値を算出し、
    　前記差の積算値が、予め設定した閾値を超えた場合に、操業異常ありと判定する請求項１に記載の高炉の異常判定方法。     The abnormality determination step is
    When a plurality of operation data to be judged are input to the stable period model, the integrated value of the difference between each input value and each output value is calculated.
    The method for determining an abnormality in a blast furnace according to claim 1, wherein when the integrated value of the difference exceeds a preset threshold value, it is determined that there is an operation abnormality.
  3. 　前記異常判定ステップは、
    　前記安定期モデルに対して判定対象となる複数の操業データを入力した際の、各入力値と各出力値との差の積算値を、正負ごとに算出し、
    　正側の差の積算値が、予め設定した正側の閾値を超えた場合、あるいは、負側の差の積算値が、予め設定した負側の閾値を超えた場合に、操業異常ありと判定する請求項１または請求項２に記載の高炉の異常判定方法。     The abnormality determination step is
    When a plurality of operation data to be judged are input to the stable period model, the integrated value of the difference between each input value and each output value is calculated for each positive and negative.
    When the integrated value of the difference on the positive side exceeds the preset positive threshold value, or when the integrated value of the negative side difference exceeds the preset negative threshold value, it is determined that there is an operation abnormality. The method for determining an abnormality in a blast furnace according to claim 1 or 2.
  4. 　前記異常判定ステップの後に、前記安定期モデルに対して判定対象となる複数の操業データを入力した際の、各入力値と各出力値との差の正負ごとの積算値を、積み重ねグラフで表示する表示ステップを更に含む請求項３に記載の高炉の異常判定方法。 After the abnormality determination step, when a plurality of operation data to be determined are input to the stable period model, the integrated value for each positive or negative difference between each input value and each output value is displayed in a stacked graph. The method for determining an abnormality in a blast furnace according to claim 3, further comprising a display step of performing.
  5. 　高炉の安定期における複数の操業データをオートエンコーダに入力することにより、入力値と出力値とが同じになるように学習させた安定期モデルを構築する学習ステップを含む安定期モデルの学習方法。 A learning method of a stable period model including a learning step to build a stable period model trained so that the input value and the output value are the same by inputting multiple operation data in the stable period of the blast furnace to the autoencoder.
  6. 　前記学習ステップは、送風流量を基準として、前記高炉の安定期における複数の操業データを選択し、前記安定期モデルを構築する請求項５に記載の安定期モデルの学習方法。 The learning step is the learning method of the stable period model according to claim 5, wherein a plurality of operation data in the stable period of the blast furnace are selected based on the air flow rate, and the stable period model is constructed.
  7. 　前記学習ステップは、前記高炉の複数の操業データのうち、前記送風流量が予め設定した閾値以上であって、かつ前記送風流量が前記閾値以上となった時点の前後の所定時間を除外した操業データを選択し、前記安定期モデルを構築する請求項６に記載の安定期モデルの学習方法。 The learning step is operation data excluding a predetermined time before and after the time when the blast flow rate is equal to or higher than a preset threshold value and the blast flow rate becomes equal to or higher than the threshold value among the plurality of operation data of the blast furnace. The learning method of the stable period model according to claim 6, wherein the above is selected and the stable period model is constructed.
  8. 　請求項１から請求項４のいずれか一項に記載の高炉の異常判定方法の判定結果に基づいて、高炉の操業を変更する高炉の操業方法。 A blast furnace operation method for changing the operation of the blast furnace based on the determination result of the blast furnace abnormality determination method according to any one of claims 1 to 4.
  9. 　高炉の安定期における複数の操業データを用いて、入力値と出力値とが同じになるように学習された安定期モデルに対して、判定対象となる複数の操業データを入力し、その際の入力値と出力値との差に基づいて、前記高炉の操業異常を判定する異常判定手段を備える高炉の異常判定装置。 Using multiple operation data in the stable period of the blast furnace, multiple operation data to be judged are input to the stable period model trained so that the input value and the output value are the same, and at that time. A blast furnace abnormality determination device including an abnormality determination means for determining an operation abnormality of the blast furnace based on a difference between an input value and an output value.
PCT/JP2021/005879 2020-03-12 2021-02-17 Abnormality determination method for blast furnace, training method for stabilization period model, operation method for blast furnace, and abnormality determination device for blast furnace WO2021182041A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP2021533830A JP7192992B2 (en) 2020-03-12 2021-02-17 Blast Furnace Abnormality Judgment Method, Stable Period Model Learning Method, Blast Furnace Operation Method, and Blast Furnace Abnormality Judgment Device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2020042856 2020-03-12
JP2020-042856 2020-03-12

Publications (1)

Publication Number Publication Date
WO2021182041A1 true WO2021182041A1 (en) 2021-09-16

Family

ID=77672201

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2021/005879 WO2021182041A1 (en) 2020-03-12 2021-02-17 Abnormality determination method for blast furnace, training method for stabilization period model, operation method for blast furnace, and abnormality determination device for blast furnace

Country Status (3)

Country Link
JP (1) JP7192992B2 (en)
TW (1) TW202143082A (en)
WO (1) WO2021182041A1 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113836821A (en) * 2021-10-26 2021-12-24 华电莱州发电有限公司 Boiler water wall tension crack online prediction method

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS4964508A (en) * 1972-10-24 1974-06-22
JPS5370852A (en) * 1976-12-07 1978-06-23 Nippon Steel Corp Display device of pressure variation patterns in blast furance
JPH0625725A (en) * 1992-07-09 1994-02-01 Nisshin Steel Co Ltd Detection of activation degree of core part in blast furnace
JP2017128805A (en) * 2016-01-19 2017-07-27 Ｊｆｅスチール株式会社 Operation method of blast furnace
KR20190078319A (en) * 2017-12-26 2019-07-04 주식회사 포스코 System and method of evaluating furnace operation state

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS4964508A (en) * 1972-10-24 1974-06-22
JPS5370852A (en) * 1976-12-07 1978-06-23 Nippon Steel Corp Display device of pressure variation patterns in blast furance
JPH0625725A (en) * 1992-07-09 1994-02-01 Nisshin Steel Co Ltd Detection of activation degree of core part in blast furnace
JP2017128805A (en) * 2016-01-19 2017-07-27 Ｊｆｅスチール株式会社 Operation method of blast furnace
KR20190078319A (en) * 2017-12-26 2019-07-04 주식회사 포스코 System and method of evaluating furnace operation state

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
AN, J: "Variational autoencoder based anomaly detection using reconstruction probability", SPECIAL LECTURE ON IE, vol. 2, no. 1, 2015, pages 1 - 18, XP055491966 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113836821A (en) * 2021-10-26 2021-12-24 华电莱州发电有限公司 Boiler water wall tension crack online prediction method
CN113836821B (en) * 2021-10-26 2023-11-28 华电莱州发电有限公司 Online prediction method for boiler water wall cracking

Also Published As

Publication number Publication date
JP7192992B2 (en) 2022-12-20
JPWO2021182041A1 (en) 2021-09-16
TW202143082A (en) 2021-11-16

Similar Documents

Publication Publication Date Title
WO2021182041A1 (en) Abnormality determination method for blast furnace, training method for stabilization period model, operation method for blast furnace, and abnormality determination device for blast furnace
CN107949813B (en) Manufacturing equipment diagnosis support device and manufacturing equipment diagnosis support method
CN104731083B (en) A kind of industrial method for diagnosing faults and application based on self-adaptive feature extraction
CN111593155A (en) Blast furnace diagnosis system and method
JP2010175446A (en) Status diagnostic apparatus
JP6750644B2 (en) Abnormality monitoring method and abnormality monitoring device
JP5875430B2 (en) Abnormality detection apparatus, program, and abnormality detection method
CN113177353A (en) Data model construction method applied to industrial early warning system
KR20220070519A (en) A production facility monitoring method, a production facility monitoring device, and a production facility operating method
JP6907663B2 (en) Extrusion load predictor for coke oven, extrusion load prediction method, computer program and computer readable storage medium
JP6686947B2 (en) Blast furnace state determination device and blast furnace operating method
JP6927461B1 (en) Operation method and operation system of production equipment
TWI745912B (en) Blast furnace abnormality determination device, blast furnace abnormality determination method and blast furnace operation method
JP2010283274A (en) Failure cause analyzer, failure cause analysis method, failure cause analysis program, and program recording medium
WO2017164368A1 (en) Monitoring device, monitoring method, and program
JP7439467B2 (en) Information processing devices, information processing systems, and model learning methods
TW202230169A (en) Abnormality diagnosing model construction method, abnormality diagnosing method, abnormality diagnosing model construction device, and abnormality diagnosing device
JP5915596B2 (en) Anomaly monitoring method and anomaly monitoring apparatus
WO2022044816A1 (en) Blast furnace condition determination device, operation method of blast furnace, and manufacturing method for molten iron
WO2024043153A1 (en) Factor selection device, factor selection method, and program
JP2696114B2 (en) Blast furnace operation management method
Wang et al. Development of monitoring and Diagnosis Technologies for steel-making process in CSC
JP6809627B1 (en) Abnormal sign detection system, abnormal sign detection device and abnormal sign detection method
JP4336262B2 (en) In-furnace pressure fluctuation prediction method and apparatus
KR20140002212A (en) Judgment method of gas distribution of blast furnace

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2021533830

Country of ref document: JP

Kind code of ref document: A

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21767496

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21767496

Country of ref document: EP

Kind code of ref document: A1